An image typically reflects the mixing of several sources, for example, lighting, reflectance, and shape. This mixing poses serious problems for a wide range of problems in image processing, computer vision, and robotics. In the first part of this talk I will discuss techniques for separating some of these image components. For example, I will show how a polarization filter and an independent component analysis can be employed to separate reflections.

In addition to this mixing, most imaging devices introduce an additional level of non-linearities (e.g., gamma correction and lens distortion). It is equally advantageous to remove these non-linearities prior to subsequent processing. I will present a technique for blindly removing these non-linearities in the absence of any calibration information or explicit knowledge of the imaging device. The basic approach exploits the fact that a non-linearity introduces specific higher-order correlations in the frequency domain (beyond second-order).

I will also talk briefly about some related work in detecting hidden messages in digital images and detecting traces of digital tampering.